{"id":"arxiv/2608.06283","url":"https://arxiv.org/abs/2608.06283","tags":["cs.LG","math.OC","math.PR","stat.ML"],"score":0,"title":"The Tamed Subgradient Unadjusted Langevin Algorithm beyond Convexity","domain":"cs.LG","source":"arxiv","authors":["Iosif Lytras","Nikolaos Makras","Sotirios Sabanis"],"reviews":0,"subject":"cs.LG","summary":"We study the problem of sampling from target distributions whose potentials are simultaneously non-smooth, subject to superlinear gradient growth, and non-convex. We introduce the Subgradient Tamed Unadjusted Langevin Algorithm (SG-TULA), a discretisation of the Langevin diffusion that operates directly on subgradients, without relying on computationally demanding smoothing procedures. To handle the superlinear regime, taming techniques are employed to produce a stable, explicit scheme. We derive non-asymptotic convergence bounds in Wasserstein-2 distance, with all constants tracked…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-08-06","reactions":{}}